Acoustic Emission Recognition Based on Spectrogram and Acoustic Features

Abstract

In order to improve the level and efficiency of fault diagnosis, an acoustic emission recognition method based on spectrum and acoustic features is proposed. The proposed system is composed of CNN and BiLSTM networks. Firstly, the amplitude spectrum and group delay phase spectrum of AE signals are extracted, and the amplitude-phase spectrum composed of the two extracted spectrum is input into CNN network to obtain the global features of AE signals. Secondly, the acoustic features such as short-term energy, zero crossing rate and kurtosis of AE signals are extracted to obtain the features of AE signals Finally, the features extracted from CNN network and BiLSTM network are fused to get the fused features, which are classified and recognized by softmax, so as to realize acoustic emission recognition. Simulation results show that the performance of the proposed system is improved by more than 17% compared with the other algorithm, and the effectiveness of the feature fusion model is verified by experiments.